The latest study, including nine inverse ETFs, is available in Bull and bear ETF backtests. The results below are archived tests of the earlier model.
These archived results predate the expanded universe. All trials and limitations of the 12-instrument version are documented on a separate page.
ETF RESEARCH / BACKTEST
Results, with the assumptions that produced them.
The SMA200 signal was tested on historical prices of actual ETFs. Review returns, drawdowns, trading activity and benchmarks together.
Calculated 10/5/2026, 12:55:16 PM JST · Model v2.0.0 · 2026 returns cover only part of the year
Win rate is profitable completed holding episodes, including trading costs and distributions, divided by all completed episodes. An episode runs from the first purchase to full liquidation and includes position adjustments. Open positions are excluded; break-even episodes are not wins. With no completed episodes, the result is unavailable (—). This is not a probability of future gains.
Shared U.S. account: archived TQQQ/SOXL backtest
2024-01-01–2026-10-02 / Standard costs / Initial capital 10,000 USD shared by TQQQ and SOXL. Each evaluation period starts fully in cash with no positions.
- Compound annual growth rate (CAGR)
- 24.39%
- Maximum drawdown
- -22.62%
- Sharpe
- 0.972
- Win rate (completed episodes)
- 40.00%
- Completed holding episodes
- 10
Total return 82.31% / Annualized volatility 26.05% / Orders 49 / Time in market 93.34%. Fewer than 30 completed episodes is an insufficient sample. Orders that adjust quantities differ from complete holding episodes. Open positions count toward ending account value but not the win-rate denominator.
- SMA200 shared U.S. account
- SMA200 shared U.S. account
| Year | Return |
|---|---|
| 2024 | 19.06% |
| 2025 | 6.09% |
| 2026 (through the evaluation end date) | 44.33% |
A multi-instrument buy-and-hold benchmark with the same budget as the shared account has not been calculated. Adding the individual results below does not produce a comparable shared-account benchmark.
FAS: model versus buy and hold
Calculated 10/6/2026, 10:15:57 AM JST. Data collection starts after the index change, on 2022-08-02. After a 220-session warm-up, evaluation begins on 2023-06-16. The October 5 bar was invalid, so this analysis ends on October 2. This is a retrospective test without parameter reselection, with few completed episodes.
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 20.42% | 6.99% | -28.13% | 0.397 | 16.67% | 6 | 23 | 86.25% |
| SMA200 with a fixed 100% target allocation | 21.76% | 7.42% | -52.48% | 0.385 | 33.33% | 6 | 12 | 86.25% |
| Leveraged ETF buy and hold | 91.91% | 26.73% | -42.32% | 0.745 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 47.69% | 15.22% | -15.24% | 0.970 | — | 0 | 1 | 100.00% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is XLF. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
TQQQ: model versus buy and hold
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 73.26% | 22.11% | -22.47% | 0.915 | 100.00% | 2 | 18 | 91.61% |
| SMA200 with a fixed 100% target allocation | 206.87% | 50.31% | -36.57% | 1.064 | 100.00% | 2 | 5 | 91.61% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
| Leveraged ETF buy and hold | 226.76% | 53.78% | -56.53% | 1.020 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 84.31% | 24.88% | -22.20% | 1.201 | — | 0 | 1 | 100.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is QQQ. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
SOXL: model versus buy and hold
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 74.13% | 22.33% | -36.41% | 0.938 | 30.00% | 10 | 30 | 79.59% |
| SMA200 with a fixed 100% target allocation | 239.26% | 55.89% | -82.67% | 0.958 | 20.00% | 10 | 21 | 79.59% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
| Leveraged ETF buy and hold | 427.22% | 82.98% | -85.54% | 1.108 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 205.99% | 50.15% | -40.08% | 1.230 | — | 0 | 1 | 100.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is SOXX. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
Other models and sensitivity diagnostics
These diagnostics have not changed the model used by the site. The ±2% band, combined rules and logistic regression are comparisons only.
| Instrument / model | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes |
|---|---|---|---|---|---|
| TQQQ / Combined rules | 4.82% | -21.95% | 0.333 | 41.18% | 17 |
| TQQQ / Logistic regression | 23.86% | -27.05% | 0.914 | — | 0 |
| TQQQ / SMA200 with a ±2% band | 17.82% | -22.47% | 0.771 | 100.00% | 2 |
| SOXL / Combined rules | 12.61% | -31.08% | 0.707 | 40.00% | 15 |
| SOXL / Logistic regression | 26.73% | -20.73% | 1.210 | 66.67% | 3 |
| SOXL / SMA200 with a ±2% band | 23.01% | -34.43% | 0.964 | 20.00% | 5 |
The diagnostics below use standard costs independently of the selector above. The 95% intervals for the Sharpe difference (comparison model minus SMA200) use 300 paired resamples of 20-session blocks. They are not significance tests of the current model against buy and hold or cash, and do not remove non-stationarity or model-search bias.
| Instrument / diagnostic | 95% interval for Sharpe difference | CAGR | Max drawdown |
|---|---|---|---|
| TQQQ / Combined rules | -1.237 – -0.072 | — | — |
| TQQQ / Logistic regression | -0.392 – 0.313 | — | — |
| TQQQ / SMA200 with a ±2% band | -0.359 – 0.018 | — | — |
| TQQQ / One-session delay | — | 18.75% | -24.18% |
| TQQQ / Pause at 20% drawdown | — | 17.13% | -27.85% |
| SOXL / Combined rules | -0.811 – 0.346 | — | — |
| SOXL / Logistic regression | -0.282 – 0.862 | — | — |
| SOXL / SMA200 with a ±2% band | -0.123 – 0.162 | — | — |
| SOXL / One-session delay | — | 19.98% | -38.98% |
| SOXL / Pause at 20% drawdown | — | 19.98% | -38.95% |
Shared U.S. account: archived TQQQ/SOXL backtest
2024-01-01–2026-10-02 / High costs / Initial capital 10,000 USD shared by TQQQ and SOXL. Each evaluation period starts fully in cash with no positions.
- Compound annual growth rate (CAGR)
- 22.60%
- Maximum drawdown
- -23.76%
- Sharpe
- 0.908
- Win rate (completed episodes)
- 30.00%
- Completed holding episodes
- 10
Total return 75.19% / Annualized volatility 26.37% / Orders 49 / Time in market 93.34%. Fewer than 30 completed episodes is an insufficient sample. Orders that adjust quantities differ from complete holding episodes. Open positions count toward ending account value but not the win-rate denominator.
- SMA200 shared U.S. account
- SMA200 shared U.S. account
| Year | Return |
|---|---|
| 2024 | 18.07% |
| 2025 | 4.25% |
| 2026 (through the evaluation end date) | 42.33% |
A multi-instrument buy-and-hold benchmark with the same budget as the shared account has not been calculated. Adding the individual results below does not produce a comparable shared-account benchmark.
FAS: model versus buy and hold
Calculated 10/6/2026, 10:15:57 AM JST. Data collection starts after the index change, on 2022-08-02. After a 220-session warm-up, evaluation begins on 2023-06-16. The October 5 bar was invalid, so this analysis ends on October 2. This is a retrospective test without parameter reselection, with few completed episodes.
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 18.87% | 6.49% | -28.89% | 0.378 | 16.67% | 6 | 24 | 86.25% |
| SMA200 with a fixed 100% target allocation | 18.43% | 6.34% | -53.08% | 0.361 | 33.33% | 6 | 12 | 86.25% |
| Leveraged ETF buy and hold | 91.66% | 26.67% | -42.36% | 0.744 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 47.44% | 15.15% | -15.27% | 0.965 | — | 0 | 1 | 100.00% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is XLF. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
TQQQ: model versus buy and hold
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 72.98% | 22.04% | -22.52% | 0.915 | 100.00% | 2 | 18 | 91.61% |
| SMA200 with a fixed 100% target allocation | 202.77% | 49.57% | -36.51% | 1.055 | 100.00% | 2 | 5 | 91.61% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
| Leveraged ETF buy and hold | 225.37% | 53.54% | -56.45% | 1.018 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 84.06% | 24.82% | -22.25% | 1.197 | — | 0 | 1 | 100.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is QQQ. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
SOXL: model versus buy and hold
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 72.55% | 21.93% | -36.89% | 0.925 | 20.00% | 10 | 30 | 79.59% |
| SMA200 with a fixed 100% target allocation | 221.38% | 52.85% | -83.44% | 0.939 | 10.00% | 10 | 21 | 79.59% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
| Leveraged ETF buy and hold | 425.64% | 82.78% | -85.51% | 1.107 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 205.75% | 50.11% | -40.15% | 1.228 | — | 0 | 1 | 100.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is SOXX. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
Other models and sensitivity diagnostics
These diagnostics have not changed the model used by the site. The ±2% band, combined rules and logistic regression are comparisons only.
| Instrument / model | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes |
|---|---|---|---|---|---|
| TQQQ / Combined rules | 3.18% | -22.71% | 0.257 | 41.18% | 17 |
| TQQQ / Logistic regression | 23.49% | -27.48% | 0.902 | — | 0 |
| TQQQ / SMA200 with a ±2% band | 17.02% | -22.52% | 0.742 | 100.00% | 2 |
| SOXL / Combined rules | 11.68% | -31.66% | 0.663 | 40.00% | 15 |
| SOXL / Logistic regression | 26.23% | -20.80% | 1.196 | 66.67% | 3 |
| SOXL / SMA200 with a ±2% band | 22.18% | -34.48% | 0.943 | 20.00% | 5 |
The diagnostics below use standard costs independently of the selector above. The 95% intervals for the Sharpe difference (comparison model minus SMA200) use 300 paired resamples of 20-session blocks. They are not significance tests of the current model against buy and hold or cash, and do not remove non-stationarity or model-search bias.
| Instrument / diagnostic | 95% interval for Sharpe difference | CAGR | Max drawdown |
|---|---|---|---|
| TQQQ / Combined rules | -1.237 – -0.072 | — | — |
| TQQQ / Logistic regression | -0.392 – 0.313 | — | — |
| TQQQ / SMA200 with a ±2% band | -0.359 – 0.018 | — | — |
| TQQQ / One-session delay | — | 18.75% | -24.18% |
| TQQQ / Pause at 20% drawdown | — | 17.13% | -27.85% |
| SOXL / Combined rules | -0.811 – 0.346 | — | — |
| SOXL / Logistic regression | -0.282 – 0.862 | — | — |
| SOXL / SMA200 with a ±2% band | -0.123 – 0.162 | — | — |
| SOXL / One-session delay | — | 19.98% | -38.98% |
| SOXL / Pause at 20% drawdown | — | 19.98% | -38.95% |
Shared U.S. account: archived TQQQ/SOXL backtest
2020-01-01–2023-12-31 / Standard costs / Initial capital 10,000 USD shared by TQQQ and SOXL. Each evaluation period starts fully in cash with no positions.
- Compound annual growth rate (CAGR)
- 17.37%
- Maximum drawdown
- -25.68%
- Sharpe
- 0.809
- Win rate (completed episodes)
- 15.79%
- Completed holding episodes
- 19
Total return 89.54% / Annualized volatility 23.12% / Orders 61 / Time in market 75.75%. Fewer than 30 completed episodes is an insufficient sample. Orders that adjust quantities differ from complete holding episodes. Open positions count toward ending account value but not the win-rate denominator.
- SMA200 shared U.S. account
- SMA200 shared U.S. account
| Year | Return |
|---|---|
| 2020 | 16.22% |
| 2021 | 45.23% |
| 2022 | -22.42% |
| 2023 | 44.75% |
A multi-instrument buy-and-hold benchmark with the same budget as the shared account has not been calculated. Adding the individual results below does not produce a comparable shared-account benchmark.
FAS: model versus buy and hold
Calculated 10/6/2026, 10:15:57 AM JST. Data collection starts after the index change, on 2022-08-02. After a 220-session warm-up, evaluation begins on 2023-06-16. The October 5 bar was invalid, so this analysis ends on October 2. This is a retrospective test without parameter reselection, with few completed episodes.
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 11.43% | 22.23% | -12.43% | 1.268 | 0.00% | 5 | 12 | 61.03% |
| SMA200 with a fixed 100% target allocation | 18.55% | 37.09% | -19.34% | 1.300 | 0.00% | 5 | 11 | 61.03% |
| Leveraged ETF buy and hold | 30.20% | 63.11% | -33.07% | 1.483 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 12.55% | 24.51% | -11.15% | 1.787 | — | 0 | 1 | 100.00% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is XLF. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
TQQQ: model versus buy and hold
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 91.38% | 17.66% | -28.87% | 0.810 | 11.11% | 9 | 33 | 72.76% |
| SMA200 with a fixed 100% target allocation | 333.95% | 44.44% | -56.68% | 0.951 | 11.11% | 9 | 19 | 72.76% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
| Leveraged ETF buy and hold | 130.99% | 23.33% | -81.53% | 0.670 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 93.21% | 17.94% | -34.58% | 0.751 | — | 0 | 1 | 100.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is QQQ. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
SOXL: model versus buy and hold
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 61.68% | 12.79% | -23.21% | 0.636 | 23.08% | 13 | 37 | 74.25% |
| SMA200 with a fixed 100% target allocation | 132.60% | 23.55% | -71.45% | 0.668 | 23.08% | 13 | 27 | 74.25% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
| Leveraged ETF buy and hold | 66.57% | 13.64% | -90.33% | 0.675 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 131.43% | 23.39% | -44.95% | 0.754 | — | 0 | 1 | 100.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is SOXX. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
Other models and sensitivity diagnostics
These diagnostics have not changed the model used by the site. The ±2% band, combined rules and logistic regression are comparisons only.
| Instrument / model | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes |
|---|---|---|---|---|---|
| TQQQ / Combined rules | 16.21% | -17.19% | 0.958 | 72.73% | 11 |
| TQQQ / Logistic regression | 9.28% | -41.54% | 0.487 | 83.33% | 6 |
| TQQQ / SMA200 with a ±2% band | 17.51% | -29.20% | 0.816 | 33.33% | 3 |
| SOXL / Combined rules | 6.02% | -25.95% | 0.412 | 37.50% | 16 |
| SOXL / Logistic regression | 6.28% | -33.12% | 0.384 | 60.00% | 5 |
| SOXL / SMA200 with a ±2% band | 14.22% | -25.55% | 0.695 | 33.33% | 6 |
The diagnostics below use standard costs independently of the selector above. The 95% intervals for the Sharpe difference (comparison model minus SMA200) use 300 paired resamples of 20-session blocks. They are not significance tests of the current model against buy and hold or cash, and do not remove non-stationarity or model-search bias.
| Instrument / diagnostic | 95% interval for Sharpe difference | CAGR | Max drawdown |
|---|---|---|---|
| TQQQ / Combined rules | -0.408 – 0.644 | — | — |
| TQQQ / Logistic regression | -0.804 – 0.286 | — | — |
| TQQQ / SMA200 with a ±2% band | -0.163 – 0.156 | — | — |
| TQQQ / One-session delay | — | 16.52% | -29.79% |
| TQQQ / Pause at 20% drawdown | — | 17.83% | -26.40% |
| SOXL / Combined rules | -0.815 – 0.411 | — | — |
| SOXL / Logistic regression | -0.803 – 0.190 | — | — |
| SOXL / SMA200 with a ±2% band | -0.123 – 0.240 | — | — |
| SOXL / One-session delay | — | 14.40% | -21.99% |
| SOXL / Pause at 20% drawdown | — | 12.82% | -23.03% |
Shared U.S. account: archived TQQQ/SOXL backtest
2020-01-01–2023-12-31 / High costs / Initial capital 10,000 USD shared by TQQQ and SOXL. Each evaluation period starts fully in cash with no positions.
- Compound annual growth rate (CAGR)
- 16.52%
- Maximum drawdown
- -26.62%
- Sharpe
- 0.776
- Win rate (completed episodes)
- 15.79%
- Completed holding episodes
- 19
Total return 84.11% / Annualized volatility 23.19% / Orders 62 / Time in market 75.75%. Fewer than 30 completed episodes is an insufficient sample. Orders that adjust quantities differ from complete holding episodes. Open positions count toward ending account value but not the win-rate denominator.
- SMA200 shared U.S. account
- SMA200 shared U.S. account
| Year | Return |
|---|---|
| 2020 | 15.83% |
| 2021 | 45.27% |
| 2022 | -23.73% |
| 2023 | 43.47% |
A multi-instrument buy-and-hold benchmark with the same budget as the shared account has not been calculated. Adding the individual results below does not produce a comparable shared-account benchmark.
FAS: model versus buy and hold
Calculated 10/6/2026, 10:15:57 AM JST. Data collection starts after the index change, on 2022-08-02. After a 220-session warm-up, evaluation begins on 2023-06-16. The October 5 bar was invalid, so this analysis ends on October 2. This is a retrospective test without parameter reselection, with few completed episodes.
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 9.68% | 18.69% | -13.15% | 1.091 | 0.00% | 5 | 12 | 61.03% |
| SMA200 with a fixed 100% target allocation | 15.52% | 30.67% | -20.41% | 1.117 | 0.00% | 5 | 11 | 61.03% |
| Leveraged ETF buy and hold | 29.76% | 62.08% | -32.95% | 1.469 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 12.26% | 23.92% | -11.15% | 1.746 | — | 0 | 1 | 100.00% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is XLF. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
TQQQ: model versus buy and hold
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 87.61% | 17.07% | -29.16% | 0.787 | 11.11% | 9 | 33 | 72.76% |
| SMA200 with a fixed 100% target allocation | 312.26% | 42.60% | -57.13% | 0.928 | 11.11% | 9 | 19 | 72.76% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
| Leveraged ETF buy and hold | 130.74% | 23.30% | -81.58% | 0.670 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 92.96% | 17.90% | -34.62% | 0.749 | — | 0 | 1 | 100.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is QQQ. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
SOXL: model versus buy and hold
Each instrument uses a separate USD 10,000 account, with comparable periods, costs and whole-share accounting. The current-model row uses standalone position sizing; it does not apply the shared account’s correlation-based scaling or investment caps.
| Method | Total return | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes | Orders | Time in market |
|---|---|---|---|---|---|---|---|---|
| SMA200 + actual ETF volatility (current) | 71.98% | 14.55% | -23.51% | 0.708 | 23.08% | 13 | 40 | 74.25% |
| SMA200 with a fixed 100% target allocation | 123.52% | 22.32% | -71.37% | 0.655 | 23.08% | 13 | 27 | 74.25% |
| Cash | 0.00% | 0.00% | 0.00% | — | — | 0 | 0 | 0.00% |
| Leveraged ETF buy and hold | 64.43% | 13.27% | -89.71% | 0.656 | — | 0 | 1 | 100.00% |
| Reference ETF buy and hold | 131.18% | 23.36% | -45.00% | 0.752 | — | 0 | 1 | 100.00% |
The fixed 100% allocation follows SMA200 direction and targets full investment when LONG. Buy and hold makes no sales after the initial purchase. The reference ETF is SOXX. Cash earns 0%. Trading costs are deducted; taxes and FX effects are excluded.
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
- SMA200 + actual ETF volatility (current)
- SMA200 with a fixed 100% target allocation
- Leveraged ETF buy and hold
- Reference ETF buy and hold
- Cash
Other models and sensitivity diagnostics
These diagnostics have not changed the model used by the site. The ±2% band, combined rules and logistic regression are comparisons only.
| Instrument / model | CAGR | Max drawdown | Sharpe | Win rate | Completed episodes |
|---|---|---|---|---|---|
| TQQQ / Combined rules | 15.23% | -17.51% | 0.912 | 72.73% | 11 |
| TQQQ / Logistic regression | 8.71% | -41.74% | 0.465 | 83.33% | 6 |
| TQQQ / SMA200 with a ±2% band | 16.73% | -29.48% | 0.783 | 33.33% | 3 |
| SOXL / Combined rules | 4.86% | -26.72% | 0.355 | 37.50% | 16 |
| SOXL / Logistic regression | 5.95% | -33.56% | 0.369 | 60.00% | 5 |
| SOXL / SMA200 with a ±2% band | 13.88% | -25.65% | 0.681 | 33.33% | 6 |
The diagnostics below use standard costs independently of the selector above. The 95% intervals for the Sharpe difference (comparison model minus SMA200) use 300 paired resamples of 20-session blocks. They are not significance tests of the current model against buy and hold or cash, and do not remove non-stationarity or model-search bias.
| Instrument / diagnostic | 95% interval for Sharpe difference | CAGR | Max drawdown |
|---|---|---|---|
| TQQQ / Combined rules | -0.408 – 0.644 | — | — |
| TQQQ / Logistic regression | -0.804 – 0.286 | — | — |
| TQQQ / SMA200 with a ±2% band | -0.163 – 0.156 | — | — |
| TQQQ / One-session delay | — | 16.52% | -29.79% |
| TQQQ / Pause at 20% drawdown | — | 17.83% | -26.40% |
| SOXL / Combined rules | -0.815 – 0.411 | — | — |
| SOXL / Logistic regression | -0.803 – 0.190 | — | — |
| SOXL / SMA200 with a ±2% band | -0.123 – 0.240 | — | — |
| SOXL / One-session delay | — | 14.40% | -21.99% |
| SOXL / Pause at 20% drawdown | — | 12.82% | -23.03% |
Daily decisions and long-term validation for the 2x Nikkei ETF (1570)
Daily signals resumed after 220 consecutive valid trading sessions were verified. The API omitted the reference ETF 1321’s 100-for-1 split on October 5, 2026; this was supplemented with cited evidence. Price/split inconsistencies in 2021 and missing observations in 2025 remain unresolved, so the long-term backtest is still withheld. Valid years are not stitched together to create a cumulative return. Research downloads retain the BLOCKED status.
Split ratio and effective date: Nomura Asset Management official disclosure. Ex-date: Monex Securities split calendar.
Backtest assumptions
- Prices and signals
- Archived daily Yahoo Finance / yfinance data, collected from 2013-01-01. Decisions use the reference ETF’s close, adjusted chronologically for splits only, and its 200-day simple moving average. Total returns including distributions are used for position sizing.
- Execution timing
- Calculated after the decision-day close plus 90 minutes. Quantities use the finalized close, with simulated execution at the next session’s open. The opening price is not used to increase order quantity. Quantities are reduced if cash is insufficient.
- Trading costs
- Standard: commission 2 bp + half-spread 3 bp + slippage 5 bp. High: 10 + 10 + 15 bp. These are one-way assumptions; 1 bp = 0.01%. Commission is applied to the execution value after spread adjustment.
- Allocation and cash
- Annualized volatility of the actual ETF over the preceding 60 sessions (at least 40 observations, floor 5%), target 25%. The shared U.S. account has an investment cap of 80%, an approximate leverage-weighted exposure cap of 150%, and a covariance-based annualized volatility cap of 25%. The remainder is cash.
- Splits and distributions
- Actual ETF prices and corporate actions are used, not synthetic prices based on three times an unleveraged index return. Distribution receivables count toward account value; cash receipt is assumed 30 calendar days after the ex-date. Taxes, settlement and cash interest are excluded.
- Metric definitions
- CAGR is the compounded annual return over calendar time. Maximum drawdown is measured from the peak, including initial capital as a potential peak. Sharpe uses 252 sessions per year, sample standard deviation and a zero risk-free rate. Annual returns are measured from the preceding year-end; partial years are not annualized.
Remaining validation work
Unresolved issues include retrospective data revisions, reconciliation of all OHLC observations with official prices, actual order books, gaps, taxes, settlement and differences in SOXL/SOXX index transitions. Forward paper trading with frozen rules, enough completed episodes, and equal-budget/equal-risk comparisons are needed. No forward paper-trading record exists yet.
Download the research data
All 240 rows of metrics by model, period and cost case, together with normalized daily equity curves used in the charts. Raw price series, real-account positions and execution ledgers are excluded. CSV returns are decimal ratios (0.1 = 10%).
In CSV files, _legacy identifies the old implementation (including comparisons that sized positions using a future opening price), _fixed denotes fixed allocation, and _proxy_vol sizes positions using reference-ETF volatility multiplied by leverage. Legacy results are not evidence for the current signal.
Archived-result identifiers
Configuration ID 486c929a24e6c0c8 / v2.0.0. SHA-256 identifies the referenced files; it does not establish data accuracy or independent audit.
- outputs/real/summary.json
- 7fff664f33b4a9c1519f72a1266ab1e849cc0e382911f124144b48427d843c8c
- config.json
- 340c1fe776012f8c12e7dd1978132f3f1ccc41a38479c4fd3014801120ee531b
- outputs/real/TQQQ/report.json
- c3a7c0a87f71b06b2f56f7f37eb85725dc566698feb6f14de33e28d6c2eba5c4
- outputs/real/SOXL/report.json
- e49345d897de5c8b9fbe9e0c83ff5a5f2d7a2b2813913dde2af4692762572e5a
- outputs/real/1570/report.json
- e62347cddbf47303e55bd5c2dd80eaa972ecb3f93bb548f016e420db7b96d12d
- FAS study: config.json
- c47470f5f69b4df1b122360dd4e986694bb06b82d5712f8668c6ae870632de5c
- FAS study: data/real/FAS.csv
- 48c2d79605b808d5db263983452775e2d2affaf79433ae4beec9ac9ec84569ba
- FAS study: data/real/FAS.json
- 01995350f960dbd7dc0d80fc2f1859ac4ef61a2b6ed3f5a02a3073e6a9c12c01
- FAS study: data/real/XLF.csv
- ee4336d491ee8ca4992f36e718caf75100a2a673ce6e9ccc8b1ebf414363a5cc
- FAS study: data/real/XLF.json
- d05fa10ff0d26c51ab869ae4664bc7754b2dbaed02286d06fd26b6bafba70960
- FAS study: lab/fas_research.py
- f5c423b3dd1305774c9700900b5f3f898edac1e5fbe62462eed4824fdf1d7ca2
- FAS study: lab/backtest.py
- ac0819bac437b89edeb23b60c2192af64a306ea5dced7268a80886b6cd911ae8
- FAS study: lab/models.py
- 2f897f79c58b21496fae3c81be76545a1724c37df9f687d9b82e63cfe5e290ed
- FAS study: lab/risk.py
- 83270af3eae5391785cdb2e9f12d65c8b18fb0da2fe71bcfe7a7e7683b306614
- FAS study: lab/execution.py
- 2a1e176419c7b1c6f438763a9cf72019ee320e884f2941a651111708cabe1791
- FAS study: outputs/fas/protocol.json
- ef62cea2d0538bdf5723f4082f2010b99060441a0b37dd5555bcb43f0876e880
Product specifications: ProShares TQQQ. Data client: yfinance documentation. Official fund performance and this research model’s performance are distinct.